The present invention relates to the field of amplifiers for audio, radio, and microwave frequencies, used in conjunction with filters, matching networks, and power combiners in systems such as transmitters, and more specifically to methods for enabling power amplifiers and their associated output networks and power combiners to operate safely and efficiently and with substantilly flat frequency responses over wider bandwidths than would otherwise be possible.
The production of high-power electronic signals is required in a variety of applications, including radio, microwave, audio, and servo amplifiers. Applications for radio and microwave power amplifiers include communications, broadcasting, and magnetic-resonance imaging. Typical applications for audio power amplifiers include communications, home entertainment, and sonar. Servo amplifiers are used for various control and positioning applications.
Electronic amplifiers are often required to operate over a wide bandwidth. An output filter is often required to prevent harmonics from reaching the load, and a matching network is often required between the load and the amplifier to provide the amplifier with a suitable load impedance. In many cases, multiple amplifiers operating in different frequency bands are employed to allow a system to operate over a large range of frequencies. It is also desirable for power amplifiers (amplifiers that produce a significant output power) to operate with high efficiency. The amplifier characteristics required to achieve these various performance goals often conflict with each other, hence designing such amplifiers using conventional techniques involves inherent trade-offs and limitations in achievable performance.
One example is an audio-frequency (AF) power amplifier (PA) driving a speaker. The resistance and inductance in the speaker form a low-pass filter, but it is desirable to maintain a flat frequency response to frequencies higher than the corner-frequency of the speaker. A second example is a radio-frequency (RF) power amplifier that drives an antenna through a matching network of limited bandwidth. The bandwidth of the matching network can prevent production of wideband signals such as spread-spectrum modulation. A third example is a modern solid-state radio transmitter that selects one output filter from a bank of filters that cover the operating frequency range of the transmitter. A relatively small filter bandwidth results in the need for a large number of filters. A fourth example is a transmitter system that employs separate HF and VHF PAs, but combines their outputs into the same antenna. A fifth example is the Meinzer modulator that uses a class-S amplifier to amplify the low-frequency components efficiently and a class-B amplifier to add the high-frequency components.
The prior art, reviewed in detail subsequently, has at least three significant limitations in these areas. Existing techniques for broadband filters and matching networks are based upon optimizing variation of gain or minimizing standing-wave ratio (SWR), but do not provide minimum voltage and current ratings for the the amplifiers that drive them. Signal-processing techniques that flatten the gain are typically based upon feedback and therefore have inherent trade-offs between the amount of gain flattening, stability, and bandwidth. Prior-art techniques for combining power amplifiers that operate on different frequencies are either lossy or require the bands to be noncontiguous.
Broadband Filters and Matching Networks
Output filters and matching networks are integral components of most amplifier systems. Filters are required to prevent the harmonics generated by power amplifiers from reaching the load. Matching networks are required because the load impedance for which the power amplifier can efficiently deliver the desired power is generally different from that of the load (e.g., antenna) into which the power must be delivered. Even small-signal amplifiers must generally be matched to the load to deliver the maximum output signal. However, the filter and matching networks impose limits on the bandwidth of the transmitter or amplifier system.
FIG. 1, FIG. 2, and FIG. 3 show an example of prior-art techniques for increasing the bandwidth. In FIG. 1, amplifier 2 drives load 9 through a series-tuned filter circuit consisting of inductor 4 and capacitor 5 with Q1=3.5 to provide adequate suppression of harmonics. As shown in the Smith-chart plot of FIG. 3, the PA-load impedance 3 exits the 2:1 SWR circle 12 at approximately 0.9041 and 1.1061 MHz, resulting in a passband ratio of 1.1061/0.9041=1.223 for a 2:1 SWR.
In this example, broadbanding is accomplished by adding a parallel-tuned circuit consisting of inductor 6 and capacitor 7 between the series-tuned circuit and the load. When the values L2 and C2 are properly chosen, the locus of the PA-load impedance 11 in FIG. 3 now loops around within the 2:1 SWR circle and exits the circle 11 at 0.8673 and 1.1529 MHz, resulting in a larger passband ratio of 1.329.
The implementation of broadband filters and matching networks is a well-known technology that is taught In textbooks by P. L. D. Abrie (The Design of Impedance-Matching Networks for Radio-Frequency and Microwave Amplifiers, Dedham, Mass., Artech House, 1985) and W. K. Chen (Broadband Matching: Theory and Implementation, Teaneck, NJ, World Scientific, 1989) and numerous articles, for example T. R. Cuthbert, Jr., xe2x80x9cBroadband impedance matching techniquesxe2x80x9d R. F. Design, vol. 17, no. 8, pp. 64-71, August 1994. A number of different methods for determining the values of the components have been developed, and the xe2x80x9creal-frequencyxe2x80x9d technique developed by H. J. Carlin (xe2x80x9cA new approach to gain-bandwidth problems,xe2x80x9d IEEE Trans. Circuits Syst.,, vol. CAS-24, no. 4, pp. 170-175, April 1977) is one of the most popular. Applications (e.g., E. Franke, xe2x80x9cSimple compensation of the single-section quarter-wave matching section,xe2x80x9d R. F. Design, vol. 15, no. 1, pp. 38-46, January 1992) and calculation programs (e.g., R. J. Dehoney, xe2x80x9cProgram synthesizes antenna matching networks for maximum bandwidth,xe2x80x9d R. F. Design, vol. 18, no. 5, pp. 74-81, May 1995) abound in the literature.
All of the existing techniques for the design of broadband filters and matching networks are based upon controlling gain (attenuation) or SWR over a bandwidth. Since the gain of a linear, lossless network and the SWR are directly related, controlling one is equivalent to controlling the other. For example, the Butterworth characteristic maximizes the bandwidth with a smoothly decreasing gain or smoothly increasing SWR. The Chebyshev characteristic minimizes the ripple in the gain or maximum SWR within a given band with a given number of elements.
All of these techniques are subject to the famous limitation discovered by Fano and described in xe2x80x9cTheoretical limitations on the broadband matching of arbitrary impedances,xe2x80x9d J. Franklin Inst., vol. 249, pp. 57-83, January 1950 and vol. 249, pp. 139-154, February 1950. The Fano limit basically states that regardless of the number of elements added to a network, it is not possible to obtain a perfect match at all frequencies and there is a fundamental trade-off between the bandwidth and the maximum gain ripple or SWR within the pass band. Thus increasing the bandwidth is achieved at the cost of increased gain ripple and increased SWR.
Delivery of power into a filter or matching network (hence the load) is, however, limited by the voltage and current ratings of the amplifier. The amplifier ratings are in turn directly related to the voltage and current that must be applied to the input of the filter or matching network to produce the desired amount of power at its output or load. In spite of the large number of publications in this area, no pre-existing techniques address the issue of of minimizing the input voltage and current so an amplifier can safely deliver maximum power to the load over a specified bandwidth.
Gain Flattening
A lossless filter or matching network implemented with real elements (inductors, capacitors) can produce a perfect match (SWR=1) at only a finite number of frequencies. The imperfect match at other frequencies (SWR greater than 1) implies a variation in gain across the passband.
The principal prior-art technologies for flattening the gain of an amplifier system are (a) manual adjustment of the signal amplitude and (b) a feedback loop. Manual adjustment has obvious limitations in speed. It is applicable only to a single-frequency signal; i.e., a signal whose bandwidth is so small that filter gain does not vary significantly over the bandwidth of the signal.
A feedback system for flattening the gain is depicted in FIG. 4. The input 21 is applied to the positive input 22 of operational amplifier 23, which then drives power amplifier 24. The output from filter 25 passes through directional coupler 26 before reaching output 27 on the load 28. Directional coupler 26 extracts a small sample of the output signal, which is applied to the negative input 29 of operational amplifier 23. As is well-known from control-system theory, variations in system gain whether due to nonlinearity or filter gain are suppressed in proportion to the open-loop gain of the system.
Feedback has been used for some time to correct gain variations. For example, Romander describes a technique for correcting distortion in the modulation process in U.S. Pat. No. 2,429,649 (1947). New variations continue to be developed. For example, K. Oosaki and Y. Akaiwa describe a system of RF feedback in xe2x80x9cNonlinearity compensation of linear power amplifier for mobile communication,xe2x80x9d Record Fourth IEEE Int. Conf. Universal Personal Commun., Tokyo, pp. 302-305, November 6-10, 1995.
A high loop gain is obviously desirable for flattening the frequency response of the system. However, maintaining stability requires that the loop gain drop to less than unity at frequencies where the phase shift is 180xc2x0 or more. Phase shifts are inherently associated with the reactive elements in the output filter, and larger numbers of elements associated with more complex filters tend to have larger phase shifts. Consequently, there is a limit to how much gain flattening can be achieved by a feedback system.
Diplexing Combiners
A number of applications require power amplifiers operating in different frequency bands to deliver power to the same load. For example, a single resistively loaded antenna may allow transmission on both HF and VHF, but component restrictions make a single PA covering both HF and VHF impractical or result in compromises in performance. In this case, one PA can be optimized for HF and another for VHF and their outputs combined by a diplexing network.
A second application is found in split-band modulators or amplifiers such as described by Meinzer in xe2x80x9cA linear transponder for amateur radio satellites,xe2x80x9d VHF Communications, vol. 7, pp. 42-57, January 1975. The envelope elimination and restoration technique developed by Kahn (Single sideband transmission by envelope elimination and restoration,xe2x80x9d Proc. IRE, vol. 40, no. 7, pp. 803-806, July 1952) achieves linear amplification of an RF signal by combining an efficient but nonlinear RF PA and a high-level amplitude modulator.. Achieving good transmitter efficiency requires an efficient amplitude modulator such as a class-S switching-mode modulator. Systems of this type are described by the inventor in xe2x80x9cHigh-efficiency single-sideband HF/VHF transmitter based upon envelope elimination and restoration,xe2x80x9d Proc. Sixth Int. Conf. HF Radio Systems and Techniques (HFxe2x80x294), York, UK, pp. 21-25, Jul. 4-7, 1994. and High-efficiency L-band Kahn-technique transmitter,xe2x80x9d Int. Microwave Symp. Digest, vol. 2, Baltimore, Md., pp. 585-588, Jun. 7-12, 1998.
The class-S modulator has, however, inherent limitation of bandwidth due to its switching frequency and output filter. The resultant limitation of the envelope bandwidth limits the bandwidth of the RF output for a given linearity as described by the inventor in (xe2x80x9cIntermodulation distortion in Kahn-technique transmitters,xe2x80x9d IEEE Trans. Microwave Theory Tech., vol. 44, no. 12, part 1, pp. 2273-2278, December 1996. The split-band modulator combines class-S and linear (e.g., class-B) amplifiers so that the class-S amplifier produces the low-frequency components of the envelope efficiently and the linear amplifier adds the high-frequency components. Since most of the power in the envelope occurs in the the low-frequency components, the efficiency of a split-band modulator is significantly higher than that of a linear modulator.
A third application occurs when two or more transmitters must use the same antenna, but each operates in its own unique frequency band. This situation occurs when two broadcast transmitters share the same tower (described, for example, by J. R. Hall in xe2x80x9cThe transmitter combiner,xe2x80x9d Communications, pp. 30-32, Mar. 4, 1970) or in a base station for cellular or PCS communication where multiple signals for multiple users must be transmitted.
Direct connection of the power amplifiers causes one to load the other, which in turn produces higher peak voltages, lower output power, lower efficiency, intermodulation distortion, and harmonics. The PAs are therefore generally coupled to the load through a combining networks called a xe2x80x9cdiplexerxe2x80x9d or xe2x80x9cmultiplexer,xe2x80x9d referred-to hereafter as a xe2x80x9cdiplexing combiner.xe2x80x9d The prior-art for diplexing combiners includes hybrid combiners, channelized combiners, circulators, and wideband networks.
Hybrid transformers are a well-known means for combining signals from two different amplifiers. Hybrid combining is described by the inventor in Section 12-7 of Solid State Radio Engineering (New York: Wiley, 1980) and is used in numerous applications, for example by Koontz in U.S. Pat. No. 5,163,181 entitled xe2x80x9cMultiple RF signal amplification method and apparatus. Hybrid combining offers the advantage of a resistive load impedance for each PA regardless of what signal is produced by the other, as well as operation of either PA at any frequency. As shown in FIG. 5, two signals 31 and 32 drive two amplifiers 33 and 34. The outputs of the two amplifiers are delivered to hybrid transformer 35, which is coupled to load 38 and dump resistor 36. When the two signals are matched in amplitude and phase, they are (ideally) combined without loss and all power is delivered to the load 38. If, however, two signals of different frequencies are combined, half of the power in each is dissipated in dump resistor 36. Consequently, the hybrid transformer is not a satisfactory means of combining signals of different frequencies when transmitter efficiency is a concern.
Channelized combiners are used when the two PAs operate on specific, different frequencies. A channelized combiner is shown in its simplest form in FIG. 6. Two signals 41 and 42 are again amplified by two separate power amplifiers 43 and 44. The amplifiers are coupled to the load 48 through band-pass filters 45 and 46. Each of these example band-pass filters comprises a single inductor (45A and 46A) and a single capacitor (45B and 46B). The first band-pass filter 45 is tuned to the frequency of signal 41, while the second band-pass filter 46 is tuned to that of signal 42. The filters pass their respective signals to the load without alteration, but present each other with a high impedance and therefore do not load each other. Each PA therefore sees a resistive load impedance (if, of course, the load is resistive) and is isolated from the signal produced by the other PA.
Channelized diplexers are described in a number of articles, for example R. Levy and K. Andersen, xe2x80x9cHF diplexer with helical resonators,xe2x80x9d Applied Microwave Mag., vol. 4, no. 2, pp. 76-87, Summer 1992, and J. R. Witmer, xe2x80x9cA modular two-band diplexer,xe2x80x9d R. F. Design, vol. 14, no. 13, pp. 30-34, December 1991. Common to all designs is, however, a significant limitation. Achieving the high out-of-band impedance so that one filter does not load the other requires that the two passbands be separated by a dead band. The channelized diplexer is therefore not suitable when continuous frequency coverage is required.
A variation on the channelized diplexer adds circulators between the filter and the antenna. Examples of this can be found in U.S. patents by Pfitzenmaier (U.S. Pat. No, 5,546,057, xe2x80x9cAntenna/filter combinerxe2x80x9d) and Piirainen (U.S. Pat. No. 5,689,219, xe2x80x9cSumming networkxe2x80x9d). The advantage of the circulator is that signals are coupled between ports only in one direction (e.g., clockwise), hence power reflected by a mismatched load is routed to a dump resistor. The fundamental limitations of the channelized diplexer remain, however. If the filters are eliminated, signals from one PA are coupled to the output of the other, which as noted above is undesirable.
The example diplexing combiner shown in FIG. 7 allows the two PAs to operate over wide ranges in frequency. Signals from sources 51 and 52 are amplified separately by PAs 53 and 54. The first PA 53 is coupled-to the load through low-pass filter 55, which comprises inductors 55A and 55C and capacitor 55B. The second PA 54 is coupled to the load through high-pass filter 56, which comprises capacitors 56A and 56C and inductor 56B. Low-pass filter 55 presents a high impedance to high-frequency signals while high-pass filter 56 presents a high impedance to low-frequency signals. Below the transition frequency, signals from PA 53 pass to the load transparently. Above the transition frequency, signals from PA 54 pass to the load transparently.
The design of broadband diplexers for signal splitting is well understood, and there is an abundance of references on this type of xe2x80x9csplitting diplexer,xe2x80x9d which is also called a xe2x80x9cbranchingxe2x80x9d or xe2x80x9cinvulnerablexe2x80x9d filter. As shown in FIG. 8, the output of a single source 61 is applied to low-pass filter 62 and high-pass filter 63. Low-pass filter delivers its output 64 to load 65, while high-pass filter delivers its output 66 to load 67.
Such a splitter is designed (as outlined, for example, by Methot in xe2x80x9cConstant impedance bandpass and diplexer filters,xe2x80x9d RF Design, vol. 9, no. 11, pp. 92-99, November 1986) by selecting the same 3-dB cut-off frequency for both filters. The components for a convenient filter characteristic such as Butterworth are the xe2x80x9c0-xcexa9xe2x80x9d values from standard tables such as those given by Zverev in Handbook of Filter Synthesis, New York: Wiley, 1967. The input admittances are conjugates, resulting in a resistive load impedance for signal source 61 at all frequencies and output gains that follow the desired filter characteristics.
Prior-art techniques such as those for designing the diplexing splitter of FIG. 8 do not apply to designing the diplexing combiner of FIG. 7. For example, xe2x80x9c0-xcexa9xe2x80x9d filter values are used and the 0-xcexa9 ends of the filters are coupled to the load, the frequency responses and PA-load impedances are erratic over a 4:1 range as shown in FIG. 9A and FIG. 9B, respectively. Different but generally similar results are obtained if the 0-xcexa9 ends of the filters are coupled to the PAs, or if filter values for resistive sources are used. Use of a splitting diplexer such as that of FIG. 7 as the signal source for the system of FIG. 8 results in a flatter system gain, but the PA-load impedances continue to vary erratically near the transition frequency, resulting in inefficient and possibly unstable operation of the PAs.
There therefore exists a need for a combining diplexer that allows continuous frequency coverage, provides flat system gain, and provides the power amplifiers with constant, resistive load impedances.
Disclosed herein is a method and apparatus for operating a system of power amplifiers, filters, matching networks, and combiners over a maximum bandwidth with minimum amplifier ratings for a given output power and with a substantially flat frequency response.
The output of an amplifier is filtered or matched to the load by a new class of xe2x80x9cminimum-ratingxe2x80x9d (MR) filter or matching network. The MR network minimizes the voltage and current required at its input for delivery to its load of a specified amount of power over a specified bandwidth. Equivalently, the MR network maximizes the bandwidth over which a specified amount of power can be delivered without exceeding specified ratings. The MR network thus overcomes a significant limitation of prior-art broadband filters, which are optimized for minimizing gain ripple or SWR over bandwidth.
The drive to an amplifier is preprocessed by a negative-component signal processor (NCSP). The NCSP is implemented through techniques such as a synthesizer and look-up table, active filter (e.g., integrated-circuit), or digital signal processor (DSP) and includes supporting devices such as A/D and D/A converters as needed. The NCSP employs negative inductors and negative capacitors whose values correspond to those of the amplifier-output filter or matching network and equivalent circuit of the load. The negative components are, however, arranged in a mirror image of those in the output filter and load. This form of signal processing has gain and phase characteristics that are exactly opposite to those of the output filter or load network. Consequently, it provides a driving signal that compensates for the effect of the output network, resulting in a substantially flat frequency response for the system.
Multiple amplifiers operating in different frequency bands are combined by driving them through an NCSP and combining their outputs through networks of complementary low- and high-pass filters or band-pass filters. The NCSP causes the signals from the amplifiers to arrive at the filters with the correct amplitudes and phases. The system frequency response is thus flat and the amplifiers are presented with matched (resistive) load impedances.
The first structural embodiment of the invention shown in FIG. 10 uses a minimum-rating network 73 to couple an amplifier 72 to the load 75. Signal source 71 provides signals within a frequency band. Load 74 may consist of resistance 74A only, or a combination of resistive and reactive elements represented here as an equivalent circuit of resistance 74A and series reactance 74B. Network 73, which may be a filter or matching network, provides minimum voltage and current ratings over the frequency band of operation. Network 73 may be fixed and optimized for a known load or adjustable for use with unknown or variable loads.
A second embodiment of the invention is the negative-component signal processor (NCSP), an example of which is shown in FIG. 11. Obtaining a flat frequency response from an amplifier or system of amplifiers may require changing the amplitude and phase of a driving signal to counteract the variations induced by an output filter. An NCSP accomplishes this by behaving as a circuit with one or more negative elements. Since true negative components do not exist, the NCSP realizes their behavior through active circuits such as operational amplifiers, a signal synthesizer with memory, or a digital signal processor.
The third structural embodiment of the invention shown in FIG. 12 preprocesses the input signal 71 with NCSP 81. The output of NCSP 81 is optionally amplified and becomes the drive to final amplifier 72, which is coupled to load 74 through filter/matching network 82. Network 82 is preferably of the minimum-rating design. The components of NCSP 72 correspond to those of load 74 and network 82, resulting in a flat frequency response from signal source 71 to output 75.
FIG. 13 illustrates a variation upon the third embodiment that is well-suited for high-efficiency amplification based upon the Kahn technique (L. R. Kahn, xe2x80x9cSingle sideband transmission by envelope elimination and restoration,xe2x80x9d Proc. IRE, vol. 40, no. 7, pp. 803-806, July 1952). Linear amplifier 72 in FIG. 12 is replaced by a nonlinear RF amplifier 77 and high-level amplitude modulator 78. NCSP 81 now provides both a phase-modulated carrier to drive RF amplifier 77 and an envelope signal to drive modulator 78. Alternatively, analog input 71 can be replaced by a digital data signal and the NCSP can generate the required waveforms directly. Doherty and outphasing power-amplifier systems can similarly be used.
A fourth embodiment of the invention (FIG. 14) adds to the second the capability to measure the characteristics of the output network and/or load and to adjust the NCSP accordingly. One method for accomplishing this is to insert a directional coupler 85 between the amplifier and output network as shown. Alternatively, the directional coupler can be inserted between the output network and the load at 87. The directional coupler samples forward- and reverse-direction signals, which are detected and optionally processed to extract impedance or reflection coefficient by detector 86. Alternately, a circuit for measurement of the amplitudes and phases of the output voltage and current can be used in place of directional coupler 85. The measurements are delivered to the NCSP for use in extraction of the equivalent circuit of the load 74 and output network 82. The extracted equivalent circuit is then used to define the negative-component equivalent circuit in the NCSP. Thus a substantially flat frequency response is maintained even under conditions when the load is unknown and variable.
The fifth structural embodiment of the invention (FIG. 15) uses multiple minimum-rating filters to combine the outputs of multiple amplifiers into the same load. Two or more input signals are produced by independent sources 241 and 242 and amplified separately by amplifiers 243 and 244. The outputs of amplifiers 243 and 244 are coupled to minimum-rating filters 245 and 246, and the outputs of the filters are coupled to common load 248 with common system output 247. The individual amplifiers operate in bands corresponding to their output filters and can thereby operate over a maximum bandwidth with minimum ratings.
In the sixth structural embodiment of the invention shown in FIG. 16, an NCSP 91 provides driving signals for two or more amplifiers 92 and 93. The outputs from the amplifiers are coupled to a common load 74 through combining network 95. The part 91A of the NCSP that preprocesses the input to amplifier 92 corresponds to the elements of output subnetwork 95A, the reactive elements 74B in load 74, and loading effects of output subnetwork 95B. Part 91B of the NCSP similarly corresponds to the elements of output subnetwork 95B, the reactive elements 74B in load 74, and loading effects of output subnetwork 95A.
The output filters can be of the minimum-rating type. When two PAs are to be combined, the output subnetworks are preferably complementary low- and high-pass filters. Because the NCSP provides driving signals with the proper amplitudes and phases, power is delivered to the load efficiently, system gain from input source 71 to output 75 on load resistance 74A is flat, and the power amplifiers are presented with resistive loads. If subnetworks 95A and 95B are of the minimum-rating type, the ratings of the PAs 92 and 93 are minimized for a given output power.
FIG. 17 shows one variation of the sixth embodiment for multiplexing three or more amplifiers. NCSP 101 provides three preprocessed driving signals to amplifiers 92, 93, and 94 which operate primarily in the frequency bands f1xe2x88x92f2, f2xe2x88x92f3, and f3xe2x88x92f4, respectively, where f1 less than f2 less than f3 less than f4. Output network 105 comprises three subnetworks 105A. 105B, and 105C whose passbands correspond to the primary operating bands of the amplifiers. The outputs of subnetworks 105A, 105B, and 105C are coupled together to form system output 106, which is coupled to load 74.
For operation at frequencies below f3, the drive to amplifier 94 is negligible and the loading of the system output 106 by subnetwork 105C is also negligible (e.g., the impedance of 105C is much higher than the load impedance 74). Consequently, amplifiers 92 and 93 and subnetworks 105A and 105B operate analogously to amplifiers 92 and 93 and subnetworks 95A and 95B in FIG. 16. For operation at frequencies above f2, the drive to amplifier 92 is negligible and the loading of the system output 106 by subnetwork 105A is also negligible. Consequently, amplifiers 92 and 93 and subnetworks 105A and 105B operate analogously to amplifiers 92 and 93 and subnetworks 95A and 95B in FIG. 16.
A second variation on the sixth embodiment for multiplexing three or more amplifiers is shown in FIG. 18. NCSP 101 again provides three preprocessed driving signals to amplifiers 92, 93, and 94 which operate primarily in the frequency bands f1xe2x88x92f2, f2xe2x88x92f3, and f3xe2x88x92f4, respectively, where f1 less than f2 less than f3 less than f4. Output network 107 comprises four subnetworks 107A, 107B, 107C and 107D. The passbands of subnetworks 107A and 107B are f1xe2x88x92f2 and f2xe2x88x92f3, while the passbands of subnetworks 107C and 107D are f1xe2x88x92f3 and f3xe2x88x92f4. At frequencies below f3, amplifiers 92 and 93 with subnetworks 107A and 107B function analogously to the system in FIG. 16. For operation over the whole band, the subsystem comprising amplifiers 92 and 93 and subnetworks 107A, 107B, and 107C is analogous to the combination of amplifier 92 and subnetwork 95A in FIG. 16. The subsystem comprising amplifier 94 and subnetwork 107D is analogous to amplifier 93 and subnetwork 95B in FIG. 16.
It will be apparent to those skilled in the art that various features of these particular embodiments can be combined to form additional embodiments. For example, the three-amplifier topologies shown in FIG. 17 and FIG. 18 are readily extended to include more than three amplifiers. The output networks for low and high bands can be coupled either in parallel as shown, or in series using transformers. The amplifiers can be of any suitable type for the frequency range and power level. Kahn or other high-efficiency amplification techniques can be used in any of the embodiments. The NCSP can be combined with other signal processing techniques suitable for the application. The diplexing and multiplexing embodiment can include load adaptation as shown in FIG. 14.